Adaptive Shared Autonomy between Human and Robot to Assist Mobile Robot Teleoperation

Mobile robot teleoperation has been widely employed when it is impractical or infeasible for humans to be present, yet still requires human judgment and decision-making skills. However, it is frustrating and stressful for human beings to merely simply teleoperate a robot without assistance due to time delay and absence of Situational Awareness (SA). On the other hand, fully autonomous robots, despite recent achievements, cannot yet execute tasks alone based on the current perception and control models. Consequently, both the human and the robot have to remain in the control loop, simultaneously contributing intelligence to the task executions, i.e. the human has to share autonomy with the robot during operation. But the challenge is how to best coordinate the two sources of intelligence from the human and the robot, to guarantee a safe and efficient task execution in remote. Therefore, a novel strategy is proposed in this thesis. It models the user intention as a contextual task to complete an action primitive, and provides appropriate motion assistance to the human operator upon the task recognition. In this way, the robot copes intelligently with the on-going tasks based on the contextual information, relieves the workload of the human operator and improves the task performance. To implement this strategy and account for the uncertainties in acquiring and processing environment information and user input, i.e. the contextual information, a probabilistic shared autonomy framework is presented to infer the contextual task the human operator performs with uncertainty measurements, and appropriately assist the human operator with the task execution according to these measurements. Since the way the human operator performs a task is implicit, it is non-trivial to model the motion pattern of the task process manually, thus a set of data-driven approaches are adopted to derive the policies of various task executions from human demonstrations, to adapt to the needs of the human operator in an intuitive way over long time. The feasibility and scalability of the proposed framework and techniques have been extensively evaluated in a variety of experiments both in simulation and on real mobile robot. With the proposed approaches, the teleoperator can be actively and appropriately assisted by increasing the cognition capability and the autonomy flexibility of the robot.